One thing everyone seems to be missing here (including the author of the article) is that medical data is an odd duck that just doesn't fit easily into a digital record. (I'm an MD, a medical informatics guy and CTO at a medical software company)
If you're running a McDonalds you can easily computerize everything: You have a fixed menu your customers can choose from, and every purchase can easily be stuffed into a relational table. Medicine isn't like that.
Trying to enter a patient encounter into a contemporary medical record system is an extremely unsatisfying experience: Humans are just weird and idiosyncratic and every time you treat someone there will be parts of the patient visit you can't represent symbolically in a piece of software. This is still largely an unsolved problem- If you read the literature on Description Logics you'll see that even PhD logicians have a hard time symbolically storing this kind of abstract data into a piece of software, let alone a doc with little computer training.
Because of this, most current record systems use a lot of "free text" for storing medical info, which is a pretty ugly hack and everyone realizes this.
I think this is a major reason for the problems people have with digital records: They don't work very well right now for fully capturing a patient encounter in a rigorous, symbolic fashion.